Automatic detection of spindles and K-complexes in sleep EEG using switching multiple models

doi:10.1016/j.bspc.2014.01.010

Biomedical Signal Processing and Control

Volume 10, March 2014, Pages 117-127

https://doi.org/10.1016/j.bspc.2014.01.010 Get rights and content

Highlights

•
We propose switching multiple models for automatic Stage 2 sleep EEG analysis.
•
This provides a unified framework for detection of multiple transient events.
•
This method is used to automatically segment EEG data and label multiple events.
•
Sleep spindles and K-complexes are successfully detected and labeled.
•
We extend the method to afford unsupervised learning of new models in real time.

Abstract

This work investigates the use of switching linear Gaussian state space models for the segmentation and automatic labelling of Stage 2 sleep EEG data characterised by spindles and K-complexes. The advantage of this approach is that it offers a unified framework of detecting multiple transient events within background EEG data. Specifically for the identification of background EEG, spindles and K-complexes, a true positive rate (false positive rate) of 76.04% (33.47%), 83.49% (47.26%) and 52.02% (7.73%) respectively was obtained on a sample by sample basis. A novel semi-supervised model allocation approach is also proposed, allowing new unknown modes to be learnt in real time.

Introduction

Sleep is categorized in five stages with each stage having unique characteristics including variations in wave patterns, eye movements and muscle tone [1]. The first four sleep stages capture the non-rapid eye movement (NREM) sleep while the fifth stage captures rapid eye movement (REM) sleep. The NREM stages are nowadays also referred to as Stages N1, N2 and N3, where N3 encompasses the previous Stage 3 and Stage 4, and the REM stage is referred to as the R stage [2]. This work focuses primarily on electroencephalographic (EEG) data recorded during Stage 2 (N2) sleep which is one of the most informative stages for an electroencephalographer searching for clinical abnormalities [3]. During this stage, EEG signals are characterized by spindles and K-complexes which according to the AASM manual of sleep scoring [2] are defined as:

Spindle A train of distinct waves having a frequency of 11–16 Hz with a duration ≥ 0.5 s, usually maximal in amplitude over central brain regions.

K-complex A well delineated negative sharp wave immediately followed by a positive component with a total duration ≥ 0.5 s, typically maximal at frontal electrodes.

Manual scoring of these two morphologically distinct waveforms which are hallmarks of Stage 2 sleep is time consuming and risks being subjectively interpreted. Thus automatic identification of these modalities would be beneficial. This has led various researchers to formulate approaches which could reliably distinguish these transient events in sleep EEG recordings. These approaches range from period-amplitude analysis [4], [5], spectral analysis through Fourier transform [6], wavelets [7] and matching pursuit [8], as well as autoregressive modelling [6]. Literature comparison of the performance obtained by different techniques is difficult because (a) dissimilar performance measures are used, (b) results are highly dependent on the manual scorers whose scoring serves as ground truth and (c) different measures are taken to reduce the problems related to human based scoring. Babadi et al. [9] for example, quote the false positive rate as the number of false alarms per second, while Devuyst et al. [5] measure this as the ratio of the number of false positives to the sum of false positives and true negatives. When different experts have labelled the same data, some authors compare results with the union of the individual scores [5] while others go through several steps to obtain a consensus scoring [10]. There is therefore no gold standard in evaluating system performance and the variance between scorers makes any measure difficult to compare objectively.

Most approaches found in the literature have been designed for the detection of either spindles or K-complexes separately. Although these approaches may be applied on the same dataset to identify both of these transient events there are no studies, to the best of our knowledge, that evaluate how the presence of one event can affect the detection of the other, considering that both are normally present in the recording. Furthermore, the approach generally taken is episodic, that is, to identify whether an epoch of the data contains one of these events. This necessitates the segmentation of the EEG recording into epochs, risking that an epoch border might occur during the transition from one event to another and thus restricting the detection of the exact transition point.

The main contribution of this work is an investigation on the use of switching multiple models [11] for the automatic detection of spindles and K-complexes in sleep EEG data. We show that this approach allows for the continuous segmentation of data which is known to switch arbitrarily and abruptly from one mode of operation to another in time, as is the case with Stage 2 sleep. Specifically we investigate the use of linear Gaussian models to represent the different modes and show that the results obtained by other authors [12], [13] on data other than EEG, where lower bounding approximations were used to handle the growing number of possible mode sequences, is generalizable to sleep EEG data. Furthermore we introduce a novel approach of extending the switching multiple model framework to a semi-supervised approach which can learn new unknown modes in real time.

The paper is organized as follows. Section 2 discusses the theory behind switching multiple models and goes into the proposed approach of identifying and learning new modes through a semi-supervised learning process. Section 3 then presents the sleep EEG data used for this analysis and the results obtained when using linear Gaussian switching multiple models for the identification of spindles and K-complexes. The identification of new modes is also discussed towards the end of this section. The paper then ends with an overall discussion and conclusion, highlighting the benefits of using the proposed switching multiple model approach for sleep EEG data segmentation.

Section snippets

Switching multiple models

Switching Multiple Models can be used for the segmentation of signals arising from systems with temporal multimodality [11], that is, systems whose dynamics change abruptly and arbitrarily in time. These models assume a divide and conquer framework where different models are made available to represent the different modes of operation. Thus the complex global behaviour of the data can be separated into a number of subdynamics which can be modelled more easily [14]. In this environment, each

Sleep EEG data

In order to investigate the performance of switching multiple models for spindle and K-complex detection, the sleep EEG data analyzed by Penny and Roberts [23] was used. This consisted of a 400 s recording from channels Fz, Cz and Pz, at a sampling frequency of 102.4 Hz. During a pre-processing stage the data was made to have zero mean and unit standard deviation. In their work, Penny and Roberts [23] applied an autoregressive Hidden Markov Model (AR-HMM) approach to detect the sleep spindles

Discussion and conclusion

This work has validated the use of an autoregressive switching multiple model for the automatic segmentation and labelling of Stage 2 sleep EEG data characterised by spindles and K-complexes. When this modelling approach was used for the detection of spindles within background EEG, quantitative results based on a sample by sample basis gave a sensitivity which ranged from 72.39% to 87.51%, depending on the expert score to which performance is being compared. This corresponded to a specificity

Acknowledgement

The authors would like to thank Dr. William Penny from University College London for providing the sleep EEG data.

References (33)

E.M. Ventouras et al.
Sleep spindle detection using artificial neural networks trained with filtered time-domain EEG: a feasibility study
Computer Methods and Programs in Biomedicine
(2005)
W.D. Penny et al.
Dynamic models for nonstationary signal segmentation
Computers and Biomedical Research
(1999)
Institute of Medicine (US) Committee on Sleep Medicine and Research
C. Iber et al.
The AASM Manual for the Scoring of Sleep and Associated Events: Rules Terminology and Technical Specifications
(2007)
J. Zeitlhofer et al.
EEG sleep spindles matched filtering
Journal of Sleep Research
(1997)
S. Devuyst et al.
Automatic sleep spindles detection – overview and development of a standard proposal assessment method
D. Gorur
Automatic Detection of Sleep Spindles
(2003)
A. Akin et al.
Detection of sleep spindles by discrete wavelet transform
U. Malinowska et al.
Micro- and macrostructure of sleep EEG
IEEE Engineering in Medicine and Biology Magazine
(2006)

B. Babadi et al.

DiBa: a data-driven Bayesian algorithm for sleep spindle detection

IEEE Transactions on Biomedical Engineering

(2012)

S.G. Fabri et al.

Functional Adaptive Control - An Intelligent Systems Approach

(2001)

P. Maybeck et al.

Reconfigurable flight control via multiple model adaptive control methods

IEEE Transactions on Aerospace and Electronic Systems

(1991)

G.S. Hoffman

A Novel Electocardiogram Segmentation Algorithm using a Multiple Model Adaptive Estimator

(2002)

S. Liehr et al.

Hidden Markov mixtures of experts with an application to EEG recordings from sleep

Theory in Biosciences

(1999)

R.A. Jacobs et al.

Adaptive mixtures of local experts

Neural Computation

(1991)

Cited by (39)

Method and system for automated detection of sleep spindles using a single EEG channels based TEO and EMD
2024, Expert Systems with Applications
As a hallmark of N2 sleep stage, sleep spindle detection based on electroencephalogram (EEG) recordings plays a crucial role in analyzing sleep. Hence, how to effectively automatically detect sleep spindle is crucially important. However, many automatic detection methods assume that the signal is stationary, which is inconsistent with the observation that spindle signals have a significant change in amplitude. Thus, some important non-stationary information and the presented evident dynamic characteristics has been ignored.
To overcome the aforementioned shortcoming, we propose a novel methodology based on Teager Energy Operator (TEO) and Empirical Mode Decomposition (EMD) to determine the exact location of sleep spindles using single-channel EEG signals. Because the TEO is sensitive to amplitude variations and energy features, our proposed method is suitable for capturing sleep spindles. We conduct plenty of experiments to evaluate our method. The experiments are conducted to validate the effectiveness of our method. On the DREAMS Sleep Spindles Database, our method achieves 91.83% $\pm$ 1.1%, 94.12% $\pm$ 3.46%, and 83.38% $\pm$ 9.32% for accuracy, specificity, and sensitivity, respectively. Experimental results show that the performance of the proposed methodology presented herein achieves better performance as compared to other methods. Moreover, owing to its usage of a single channel of EEG signal, the proposed method will be suitable for edge-device implementation to expedite sleep disorder diagnosis.
Optimizing sleep staging on multimodal time series: Leveraging borderline synthetic minority oversampling technique and supervised convolutional contrastive learning
2023, Computers in Biology and Medicine
Sleep is an important research area in nutritional medicine that plays a crucial role in human physical and mental health restoration. It can influence diet, metabolism, and hormone regulation, which can affect overall health and well-being. As an essential tool in the sleep study, the sleep stage classification provides a parsing of sleep architecture and a comprehensive understanding of sleep patterns to identify sleep disorders and facilitate the formulation of targeted sleep interventions. However, the class imbalance issue is typically salient in sleep datasets, which severely affects classification performances. To address this issue and to extract optimal multimodal features of EEG, EOG, and EMG that can improve the accuracy of sleep stage classification, a Borderline Synthetic Minority Oversampling Technique (B-SMOTE)-Based Supervised Convolutional Contrastive Learning (BST-SCCL) is proposed, which can avoid the risk of data mismatch between various sleep knowledge domains (varying health conditions and annotation rules) and strengthening learning characteristics of the N1 stage from the pair-wise segments comparison strategy. The lightweight residual network architecture with a novel truncated cross-entropy loss function is designed to accommodate multimodal time series and boost the training speed and performance stability. The proposed model has been validated on four well-known public sleep datasets (Sleep-EDF-20, Sleep-EDF-78, ISRUC-1, and ISRUC-3) and its superior performance (overall accuracy of 91.31–92.34%, MF1 of 88.21–90.08%, and Cohen’s Kappa coefficient $k$ of 0.87–0.89) has further demonstrated its effectiveness. It shows the great potential of contrastive learning for cross-domain knowledge interaction in precision medicine.
Detection of k-complexes in EEG signals using a multi-domain feature extraction coupled with a least square support vector machine classifier
2021, Neuroscience Research
Sleep scoring is one of the primary tasks for the classification of sleep stages using electroencephalogram (EEG) signals. It is one of the most important diagnostic methods in sleep research and must be carried out with a high degree of accuracy because any errors in the scoring in the patient’s sleep EEG recordings can cause serious problems. The aim of this research is to develop a new automatic method for detecting the most important characteristics in sleep stage 2 such as k-complexes based on multi-domain features. In this study, each EEG signal is divided into a set of segments using a sliding window technique. Based on extensive experiments during the training phase, the size of the sliding window is set to 0.5 s (s). Then a set of statistical, fractal, frequency and non-linear features are extracted from each epoch based on the time domain, Katz's algorithm, power spectrum density (PSD) and tunable Q-factor wavelet transform (TQWT). As a result, a vector of twenty-two features is obtained to represent each EEG segment. In order to detect k-complexes, the extracted features were analysed for their ability to detect the k-complex waveforms. Based on the analysis of the features, twelve out of twenty-two features are selected and forwarded to a least square support vector machine (LS-SVM) classifier to identify k-complexes in EEG signals. A set of various classification techniques of K-means and extreme learning machine classifiers are used to compare the obtained results and to evaluate the performance of the proposed method.The experimental results showed that the proposed method, based on multi-domain features, achieved better recognition results than other methods and classifiers. An average accuracy, sensitivity and specificity of 97.7 %, 97 %, and 94.2 % were obtained, respectively, with the CZ-A1 channel according to the R&K standard. The experimental results with high classification performance demonstrated that the technique can help doctors optimize the diagnosis and treatment of sleep disorders.
Multitaper-based method for automatic k-complex detection in human sleep EEG
2020, Expert Systems with Applications
Citation Excerpt :
And, the effort required from human experts to analyze EEG recordings visually and detect KC occurrences is a costly, time-consuming and error-prone task (Younes, 2017). Recent research works have proposed automatic methods for KC detection (Camilleri, Camilleri, & Fabri, 2014; Devuyst, Dutoit, Stenuit, & Kerkhofs, 2010; Erdamar, Duman, & Yetkin, 2012; Krohne, Hansen, Christensen, Sorensen, & Jennum, 2014; Lajnef et al., 2015; Parekh, Selesnick, Rapoport, & Ayappa, 2015; Patti et al., 2016; Yazdani, Fallet, & Vesin, 2018; Yücelbaş et al., 2018), varying greatly in approach and performance. In general, these automatic methods reach mean F1 and F2 values (harmonic averages of precision and recall) inferior to 75%, configuring promising approaches but in need of improvement.
In this paper, we propose a novel method for automatic k-complex (KC) detection in human sleep EEG, named MT-KCD. KCs are slow oscillations in the EEG signal characterized by a well-delineated, negative, sharp waves immediately followed by a positive component standing out from the background, with high-amplitude and total duration ≥ 0.5 s. Among the important aspects of the KC are its homeostatic and reactive functions in the brain, functioning as a sleep protection mechanism, and its practical use as a marker of N2 sleep stage during sleep studies. Given the importance of the KC, and the effort required from human experts to analyze EEG recordings visually, some recent research works have proposed automatic methods for KC detection. In comparison with existing methods, a key feature and novelty of MT-KCD is the use of multitaper spectral analysis to pre-process the EEG signal and automatically extract candidate KCs from it (characterized as 0-4 Hz power concentrations standing out from the background). After extraction, candidates are accepted/rejected depending on time domain characteristics (peak-to-peak amplitude ≥ 75 µV, duration ≤ 2 s). The method overall time complexity is $O (N \cdot \log N)$ . Regarding effectiveness, we have evaluated MT-KCD by using a public KC database (DREAMS) consisting of ten polysomnographic recordings of healthy patients (6 female and 4 male subjects with age range 20–47 years) partially annotated by two experts. Results have shown that MT-KCD improves detection metrics, especially F1 and F2 scores (harmonic averages of recall and precision), when compared to existing methods. Besides, improving F1 and F2 scores, MT-KCD also contributes to the automatic analysis of sleep EEG multitaper spectrograms, a technique recently proposed by researchers in the area of sleep studies as a complement to the traditional hypnogram (sleep stages diagram).
A fuzzy neural network approach for automatic K-complex detection in sleep EEG signal
2018, Pattern Recognition Letters
Citation Excerpt :
Recently, remarkable progress have been made to find these stages of sleep with the help of the computer. However, the automatic detection of stages of sleep using conventional methods is inadequate due to the short duration sleep phenomena of the EEG, such as K-complexes and sleep spindles that are trait for light sleep (stage 2) [5,11,38,48]. Conversely, the application of smart systems like neural networks and fuzzy logic unveils a safe, accurate and reliable detection and pattern identification of the signal.
The study of sleep stages and the associated signals have emerged as a very important parameter to identify the neurological disorders and test of mental activities nowadays. Electroencephalogram (EEG) is an electrophysiological method for monitoring, managing, and diagnosing the mental disorders or neurological problems. The EEG signals are highly transient and nonlinear in nature. It varies with the mental conditions. In the sleep state, a non-stationary wave generates with comparatively higher peaks is known as K-complex. The K-complex is a kind of transient wave which can be seen in the NREM stage II sleep. The main difficulty behind the design of the automated K-complex detection system is a nonlinear and dynamic characterization of it. The other difficulty for the system design is the very much similar behaviour of K-complex to other EEG wave. To overcome these problems, in this paper we are giving the detailed description for developing an automatic K-complex detector using fuzzy neural network approach. In this method, fuzzy C-means algorithm is utilized for the rough and rapid recognition of K-complex and the neural network classifier does the exact evaluation on the detected K-complex. One more fast computing Back Proportion algorithm is used for train the network in this work. This technique of detection of K-complex with a well-known pattern present in sleep EEG is a fuzzy neural based software solution in the field of biomedical signal processing.
A single channel sleep-spindle detector based on multivariate classification of EEG epochs: MUSSDET
2018, Journal of Neuroscience Methods
Citation Excerpt :
Improving the automatic recognition of individual sleep stages with the help of automatic sleep-spindle detections. Analyzing the effect of various neurological diseases such as Parkinson’s disease and epilepsy on sleep quality(Carrion et al., 2010; Stavitsky and Cronin-Golomb, 2011). The assessment of the developed detectors, which was described in detail and with consideration of recommendations presented by previous studies, constitutes an important property of our work.
Studies on sleep-spindles are typically based on visual-marks performed by experts, however this process is time consuming and presents a low inter-expert agreement, causing the data to be limited in quantity and prone to bias. An automatic detector would tackle these issues by generating large amounts of objectively marked data.
Our goal was to develop a sensitive, precise and robust sleep-spindle detection method. Emphasis has been placed on achieving a consistent performance across heterogeneous recordings and without the need for further parameter fine tuning. The developed detector runs on a single channel and is based on multivariate classification using a support vector machine. Scalp-electroencephalogram recordings were segmented into epochs which were then characterized by a selection of relevant and non-redundant features. The training and validation data came from the Medical Center-University of Freiburg, the test data consisted of 27 records coming from 2 public databases.
Using a sample based assessment, 53% sensitivity, 37% precision and 96% specificity was achieved on the DREAMS database. On the MASS database, 77% sensitivity, 46% precision and 96% specificity was achieved. The developed detector performed favorably when compared to previous detectors. The classification of normalized EEG epochs in a multidimensional space, as well as the use of a validation set, allowed to objectively define a single detection threshold for all databases and participants.
The use of the developed tool will allow increasing the data-size and statistical significance of research studies on the role of sleep-spindles.

View all citing articles on Scopus

View full text

Automatic detection of spindles and K-complexes in sleep EEG using switching multiple models

Highlights

Abstract

Introduction

Section snippets

Switching multiple models

Sleep EEG data

Discussion and conclusion

Acknowledgement

Computer Methods and Programs in Biomedicine

Computers and Biomedical Research

Institute of Medicine (US) Committee on Sleep Medicine and Research

The AASM Manual for the Scoring of Sleep and Associated Events: Rules Terminology and Technical Specifications

EEG sleep spindles matched filtering

Journal of Sleep Research

Automatic sleep spindles detection – overview and development of a standard proposal assessment method

Automatic Detection of Sleep Spindles

Detection of sleep spindles by discrete wavelet transform

Micro- and macrostructure of sleep EEG

IEEE Engineering in Medicine and Biology Magazine